A method for tracking and trajectory prediction of a UAV based on radar plot information
By using adaptive spatiotemporal coding and physical constraint feature extraction based on radar point information, combined with Transformer and Kalman filtering, the problem of tracking and predicting low, slow, and small targets in complex environments is solved. This achieves high-precision, robust trajectory prediction and system adaptability, making it suitable for embedded systems.
Patent Information
- Authority / Receiving Office
- CN · China
- Patent Type
- Patents(China)
- Current Assignee / Owner
- CHONGQING UNIV
- Filing Date
- 2026-03-30
- Publication Date
- 2026-07-07
AI Technical Summary
Existing technologies for tracking and predicting the trajectory of low, slow, and small targets in the low-altitude airspace suffer from poor adaptability to complex maneuvers, fixed noise parameters, and a lack of physical consistency in deep learning methods. In particular, under low signal-to-noise ratio, high maneuverability, and sparse observation environments, it is difficult to achieve stable and accurate state estimation and trajectory prediction.
A radar-based approach is employed, employing adaptive spatiotemporal coding, physical constraint feature extraction, Transformer-based spatiotemporal hybrid prediction, and adaptive Kalman filter correction to construct a data-driven pattern learning and model-driven estimation cascade framework. By combining causal convolutional neural networks and Kalman filtering, semantic understanding of target motion patterns and embedding constraints of physical laws are achieved.
It improves tracking accuracy and robustness in complex environments, ensures the physical reliability of the trajectory and the adaptability of the system, and is suitable for embedded systems with limited computing resources, meeting the requirements of real-time performance and reliability.
Smart Images

Figure CN121959458B_ABST
Abstract
Citation Information
Patent Citations
Maneuvering target intelligent tracking method based on self-attention mechanism and physical constraint
CN120065200A
Hybrid computing architectures with specialized processors to encode / decode latent representations for controlling dynamic mechanical systems
US20220314434A1